Jacob Coxon quit Anthropic and the industry. Hubinger put better than one-in-ten odds AI kills everyone this decade

Jacob Coxon, 27, a pretraining researcher who spent three years at OpenAI and then Anthropic, announced his resignation on X Tuesday and said he is leaving the industry altogether, The Next Web writes. Pretraining is the big early training run that teaches a model general skills before later fine-tuning.

Coxon said neither company is acting responsibly. He said both are racing straight to self-improving superintelligence and gambling with our lives. Self-improving systems are ones that can make themselves smarter without waiting on human researchers.

He called safety work inadequate under competitive pressure, not fake. Researchers see the hazard and continue because a competitor will move if they stop.

Developing this class of capability inside private companies, he argued, should not be those companies' alone.

The next day, Evan Hubinger, Anthropic's Alignment Science lead, answered him in public. Alignment is work meant to keep powerful AI systems doing what humans intend.

“Jacob is correct here; we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” Hubinger wrote. He added that Anthropic is trying its best, but that the company does not yet have a plan to solve alignment for superintelligence and is not clearly on track to.

Hubinger's figure is a personal estimate, not Anthropic's official risk number. Subjective probabilities on unprecedented events are not measurements. Other researchers put the figure far lower or near zero.

Anthropic had not issued a corporate response to Coxon's resignation as of the TNW piece.

Axios frames the same morning as OpenAI begging for someone to slow the AI race. That verb is Axios's framing of OpenAI's public posture, not an OpenAI press release. The piece says executives and researchers increasingly see a race they cannot safely slow on their own, and urge governments, rivals, and outside institutions to impose restraint.

Axios notes Coxon's exit from the OpenAI-Anthropic race toward hard-to-control systems, and places it against a week of OpenAI capability unveilings. Dean Ball, OpenAI's head of strategic futures, published a personal essay on “self-sovereign” AI agents beyond human control.

OpenAI says it is working on safeguards governments have yet to require, including a formal policy for publicly reporting serious AI incidents. The Trump administration largely resists binding AI regulation and treats slowing as its own threat.

The overnight window now has three public documents that belong together. A named trainer walks. An alignment lead puts a personal extinction percentage on the record and admits there is no plan.

A major outlet reads the frontier as asking someone else to brake. That is still not a corporate slowdown. The people closest to the runs are writing the liability memo in public while the companies keep racing.

Sources: The Next Web; Axios.